For this estimation of an Engel Curve, I wanted to use a good that was not clearly a normal or inferior good. For this purpose, I used monthly Honda Civic sales since 2010 as my dependent variable. To estimate income, I used FRED’s data on real personal disposable income per capita in 2009 dollars. Since this was a time series, I went in anticipating autocorrelation, so I included a Durbin-Watson test to detect. With all the data gathered, I ran a regression in SAS with a Durbin-Watson test.
For this sample, a $1 increase in real personal disposable income per capita was associated with 2.66 additional Honda Civics sold a month. Both the coefficient on the income variable and the regression as a whole were statistically significant (p<.0001). Despite this, the coefficient of determination was 0.2802 indicating a weak fit.
The results of the Durbin-Watson test resulted in a P-value <.0001 indicating likely autocorrelation. To correct for this autocorrelation I used the AutoReg procedure in SAS.
For the corrected sample, a $1 increase in real personal disposable income per capita was associated with 3.01 additional Honda Civics sold a month instead of 2.66 additional Civics sold. The coefficient was still statistically significant with a P value of 0.0001. The regression still fit poorly.
These regressions do seem to indicate that Honda Civics are a normal good, but the poor fit indicates that data over a longer stretch of time may be helpful in generating a more convincing regression.